The main problem with academic research process, I think, is the notion that a publication is the end goal of research. This is true only for theory and math research. In systems research, a working system should always be the end goal of research; a publication merely serves as documentation of the working system.

Scientific publishing is important. Hadoop was possible largely because Google published SOSP papers on map-reduce and google file system. Many of the current NoSQL stores are based on ideas from the SOSP paper on Dynamo key-value store from Amazon. However, it is important to realize that Dynamo, Google File System, Map Reduce were all working systems first and publications later.

Unfortunately, there is no incentive built into the academic research process today for building working systems. It is not necessary for promotion/tenure/awards. The incentive structure for systems research is what really needs to change.